摘要
地震前兆台网观测数据异常图像识别方法一直是地震监测预报人员研究的重要课题。为提高异常图像识别的工作效率,充分利用已有的异常图像识别经验知识,开展基于卷积神经网络(CNN)的快速异常识别方法探索性研究。结果表明:基于CNN的异常图像识别方法准确率较高,实现了异常图像的快速识别。整个台网的异常图像丰富多样,影响较多。由于特定观测手段下,特定影响因素的训练样本少,该方法应用于整个台网的异常图像的自动识别,还需要进一步开展研究工作。
How to recognize anomaly images from the observation data of earthquake precursory networks is always an important topic for earthquake monitoring and forecasting personnel.In order to improve the work efficiency of abnormal image recognition and make full use of the existing experience and knowledge of abnormal image recognition,an exploratory research on the rapid CNN-based identification of anomaly images was carried out.The results show that the CNN-based abnormal image recognition method proposed in this paper has high accuracy,and can realize the rapid recognition of abnormal images.Due to the lack of training samples with specific observation means and specific influencing factors,further research is needed to apply this method to automatic recognition of abnormal images in the whole network.
作者
王军
刘春国
樊俊屹
WANG Jun;LIU Chunguo;FAN Junyi(China Earthquake Networks Center, Beijing 100045, China)
出处
《地震工程学报》
CSCD
北大核心
2021年第1期28-32,49,共6页
China Earthquake Engineering Journal
基金
国家自然科学基金面上项目(41372349)
中国地震局专项“全国地球物理台网数据跟踪分析与产出”。
关键词
卷积神经网络
地震前兆台网
异常图像
自动识别
convolutional neural network
earthquake precursor observation networks
abnormal image
automatic identification
